Knowledge Discovery Of Game Design Features By Mining

5Marketing is another field where gamification concepts have been successfully implemented in order to induceengagement, brand loyalty and brand awareness. These three key marketing concepts are relevant in the gamification context:engagement – “high relevance of brands to consumers and the development of an emotional connection between consumers andbrands” (Rappaport, 2007), brand loyalty – “the relationship between relative attitude and repeat patronage” (Dick and Basu,1994) and brand awareness, “the rudimentary level of brand knowledge involving, at the least, recognition of the brand name”(Hoyer and Brown, 1990). There has also been work that shows the positive attitude of marketing executives towards adoptinggamification in order to improve the above mentioned three marketing concepts (Lucassen and Jansen, 2014).Literature relevant to gamification in wide range of fields is summarized in Table 1.Table 1: Gamification Applications in Various DomainsDomainRelevant LiteratureEducation/(Cheong et al., 2013b; Denny, 2013; Dong et al., 2012; Fitz-Walter et al., 2011; Foster et al., 2012; Goehle, 2013; Li et al., 2012)LearningHealth and Wellness(Cafazzo et al., 2012; Hamari and Koivisto, 2013; Hori et al., 2013; K. Rose et al., 2013; Stinson et al., 2013)Marketing(Hamari and Järvinen, 2011)Sustainability(Berengueres et al., 2013; Gnauk et al., 2012; Gustafsson et al., 2009)Commerce(Hamari, 2013)Crowd Sourcing(Liu et al., 2011)2.2 Game Design FeaturesAccording to Werbach and Hunter, there are three categories of game features that are relevant to gamification:Dynamics, Mechanics and Components. (Werbach and Hunter, 2012). The authors define the three features as follows: “Dynamics” are the big-picture aspects of the gamified systems that you have to consider and manage but which you neverdirectly enter into the game. Analogies in the management world would be employee development, creating an innovativeculture, etc. “Mechanics” are the basic processes that drive the action forward and generate player engagement” “Components” are the specific instantiations of mechanics or dynamics”

6Table 2.1 and 2.2 enlist game mechanics and components defined by them (Werbach and Hunter, 2012) and the relevantliterature for the successful implementation of each game feature. Table 2.3 presents the mechanics of latent gamification featuresand their corresponding components. Availability of a wide array of game design features makes it challenging for designers toincorporate all game design features into a single application or game. Though there has been a modest amount of research inevaluating the success of specific game design features, the relationship between successful games and the game design featuresthat they contain, remains an open research question. This paper will focus on identifying the game design features that contributetowards the success of a game. In this paper, mechanics and components are explored, given that dynamics does not directly enterinto the design of a game and is more abstract in nature. Knowledge gained from this work will enable designers to incorporategamification strategies into their decision making processes in order to motivate and enhance user engagement.Table 2.1: Literature of Various Game Design FeaturesMechanicsGame Design FeaturesRelevant LiteratureChallenges – Puzzles or other tasks that require effort to solve(Domínguez et al., 2013; Dong et al., 2012; Flatla et al., 2011)Feedback – Information about how the player is doing(Dong et al., 2012; Gustafsson et al., 2010; Li et al., 2012)Rewards – Some benefits that go together for some action or achievement in the(Downes-Le Guin et al., 2012; Liu et al., 2011; Li et al., 2012)gameTable 2.2: Literature of Various Game Design FeaturesComponentsGame Design FeaturesRelevant LiteratureAchievements – A form of reward attached to performing specific actions(Fitz-Walter et al., 2011; Liu et al., 2011; Montola et al., 2009)(Berengueres et al., 2013; Downes-Le Guin et al., 2012; Liu et al.,Avatars – Visual representations of players’ characters2011; K. Rose et al., 2013)(Anderson et al., 2013; Denny, 2013; Domínguez et al., 2013;Badges – Visual representations of achievementsHakulinen et al., 2013)Domínguez et al., 2013; Farzan et al., 2008; Gnauk et al., 2012; HalanLeaderboards – Visual displays of player progression and achievementset al., 2010Levels – Defined steps in player progressionDomínguez et al., 2013; Dong et al., 2012; Farzan et al., 2008

7Points – Numerical representation of game progression(Farzan et al., 2008b; Halan et al., 2010)Social graph – Ability to track progress of friend and enables interaction(Hamari and Koivisto, 2013; Shi et al., 2014; Simões et al., 2013)Table 2.3 Latent Game Design FeaturesMechanicsComponentsBoss Fights – Especially hard challenges at the culminationChance – Involvement of luck from a random mechanismof a levelCompetition – Getting players to compete against one anotherCollections – Set of items or badges to accumulateContent unlocking - Unlocks new levels/new features whenCooperation – Getting players to work together to achieve a shared goalplayers reach specific objectivesGifting – Gives an opportunity to gift things such asResource acquisition – Obtaining useful or collectible itemlives/points to other playersTransactions – Buying, selling or trading with other human players or automated playersQuests – Predefined challenges with objectives and rewardsTurns – Sequential participation by alternating playersTeams – Defined group of players working towards acommon goalWin states – The state that defines winning the gameVirtual Goods – game assets with perceived or real moneyvalue3. MethodologyThis work seeks to discover whether there exists a set of game design features that are common across successful task drivenapplications. Figure 2 presents an outline of the proposed methodology that includes Data Sampling (3.1), Data Collection (3.2) andModel Generation and Validation (3.3).

8Figure 2: Outline of the Methodology3.1 Data Sampling3.1.1 Identification of ‘m’ game design featuresGame design features are defined as the building blocks or features shared by games and not just mere elements that are necessaryfor building games (Deterding et al., 2011). In the methodology presented in Figure 2, ‘m’ game design features are identified usingpreviously conducted research on gamification and game design. Research on game design features in the past has led to the discoveryof a set of game design features. However, there is a knowledge gap between the existence of these game design features and theirimpact on the success/failure of games. An exhaustive list of game design features is given in Tables 2.1, 2.2 and 2.3. A total of 24game design features are identified from literature. Due to continuous integration of newly developed features into games, it ispossible that the game design features considered in this work are a subset of the total existing game design features that will emergein the future. The 24 game design features are classified into two main categories: Mechanics and Components. In this work gamedesign features are considered individual entities building a game, in order to avoid preconceived classifications of their functions.3.1.2 Sampling of ‘n’ games to be studiedTo evaluate the impact of extracted game design features, it is essential to identify games that can be played using various gamingplatforms such as mobile, PCs, Consoles, etc. Mobile games are becoming ubiquitous in today’s society. The number of smartphoneusers has increased exponentially in the last few years with an increase of over 100 million in a period of one year from 2012 to2013. Statistics reports in 1990s suggested that 78% of adults in US own a smartphone and according to recent estimates total numberof smartphone users was 1.75 billion is 2014 which is expected to grow up to 2.50 billion in 2017 worldwide(Johnson and Maltz,

91996; Munoz et al., 2015). Moreover, a higher percentage of users spend time downloading applications from the platform marketsfor iOS and Android (Feijoo et al., 2012; Lee, 2012). Online game or app downloading websites have become popular in recent yearsdue to scalability and ease of use (Purcell, 2011). In this work, games to be investigated are selected based on their ranking in theplatform application market. The rank of a game is indicative of its success, compared to other games in the pool (Filho et al., 2014).In an attempt to reduce bias from the end result, it is necessary to select games that are highly successful or ranked and games thatare highly unsuccessful. Sampling games in this manner will enable researchers to identify the combination of game features thathave an impact on the success as well as failure of a game. The sample games to be studied can either be selected randomly or basedon the genre (arcade, action, trivia, etc.) and then classified as successful or unsuccessful based on their respective rankings withinthe gaming platform market.3.2 Data CollectionAfter the identification of game design features and games to be sampled based on the rankings of the games in the gaming market,an ‘n x m’ binary input matrix is constructed based on the presence or absence of each game design feature (selected in section 3.1.1)on the level of success or failure of a game (selected in section 3.1.2). The binary matrix gets input by analyzing each game for thepresence or absence of the game design features identified.Table 3 Representation of Binary Matrix – State of GameGameGame Design Feature 1Game Design Feature 2 .Game Design Feature ‘m’State of GameGame 1Game 2.Game ‘n’PresentAbsent.AbsentAbsentAbsent.Present . . PresentPresent . . .PresentPositive OutcomeNegative Outcome . . .Positive OutcomeTable 3 represents the binary matrix constructed using inputs on game design features and the output of the state of the game. Thestate of the game in Table 3, represents whether a game was successful or unsuccessful. A successful game will be a positive outcomeand an unsuccessful game will be a negative outcome. The success of a game is based on its ranking in the gaming market (Filho etal., 2014). A higher ranked game is successful, as rankings are governed by the number of users and ratings for the game. Similarlya significantly lower ranked game will be categorized as unsuccessful.

103.3 Model GenerationIt has been discovered that the market demand of a product is positively correlated with customers’ feedback expressed on largescale digital platforms (Tuarob and Tucker, 2015, 2013). It is essential that such feedback is taken into account while designing newproducts or advanced versions of existing products. Users provide feedback about their experience with a product through ratings.These ratings are aggregated to score the product based on the users’ response and ranked in comparison to other products within itsdomain. Filho et al. claimed that rankings of a game are directly related to their success in the gaming market (Filho et al., 2014).Based on the game design features existing in a game, and the success/failure outcome of a game (based on users’ feedback measuredby the ranking of a game), the impact of i) individual game design features and ii) combinations of game design features on thesuccess/failure of a game can be quantified.3.3.1 Model Generation using Individual FeaturesThe relationship between the presence/absence of the game design feature and the success/failure of the game, is quantified basedon a confusion matrix. Table 4 illustrates a confusion matrix, where the columns represents the presence and absence of a particulargame design feature and the rows represents the success/positive outcome and failure/negative outcome of a game, determined basedon its ranking. Performance of a classification is evaluated using a confusion matrix, with measures such as precision, recall and FScore, employed to evaluate the robustness of a classification (Buckland and Gey, 1994). In this work, precision (Powers, 2011)represents the ratio of the number of successful games that have a game design feature to the total number successful games withand without that particular game design feature usually expressed as percentage (equation 1).𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑇𝑟𝑢𝑒 � 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒Recall (Powers, 2011) is the ratio of the number of successful games that have a game design feature, to the total number successfuland unsuccessful games with that particular game design feature (equation 2).𝑅𝑒𝑐𝑎𝑙𝑙 𝑇𝑟𝑢𝑒 � 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝐹𝑎𝑙𝑠𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒Table 4 Representation of Confusion MatrixGame Design FeaturePositive OutcomePresentTrue Positive(TP)AbsentFalse Positive(FP)

11ClassVariableNegative OutcomeFalse Negative(FN)True Negative(TN)The confusion matrix in Table 4 shows the presence or absence of a game design feature and the categorized state of a game (i.e.,positive or negative), based on its rankings. A confusion matrix defines the relation between the true condition (i.e., the presence ofa given game design feature) and the predicted condition (i.e., the positive or negative outcome of a game). In this case, True Positive(TP) will be defined as “positive outcome for a game when a game design feature is present”. False Negative (FN) will be definedas “negative outcome for a game when a game design feature is present”. False Positive (FP) will be defined as “positive outcomefor a game when a game design feature is absent” and True Negative (TN) will be defined as “negative outcome for a game when agame design feature is absent”.The F-Score is a metric that characterizes the combined performance of both precision and recall. It is the harmonic meanof precision and recall (equation 3). The F-Score provides a statistical measure of the agreement between the ground truth and classvariable. The F-Score enables the assessment of a classification algorithm’s ability to correctly distinguish between features relevantto a class variable and features irrelevant to a class variable (Huang et al., 2005). An F-Score of 1.0 means that a particular gamedesign feature consistently predicts the success/failure of a game (i.e., with no type 1 or type 2 error). Literature suggests that the FScore is a composite measure that favors algorithms with higher sensitivity and specificity. Classical measures such as precision,recall and F-Score are frequently used to evaluate the capabilities of algorithms (Sokolova et al., 2006) and will therefore beemployed in this work to assess the veracity of the game design feature classification models.𝐹 𝑆𝑐𝑜𝑟𝑒 2 (𝑃𝑟𝑒𝑐𝑖𝑠𝑜𝑛 ��𝑜𝑛 𝑅𝑒𝑐𝑎𝑙𝑙)(3)3.3.2. Model Generation using Combination of Game Design FeaturesWhile individual gamification features may impact the success or failure of a game, there may be unique game design featurecombinations that are better predictors of a game’s success or failure. Exploring game design feature combinations requires morecomplex mathematical approaches beyond single feature confusion matrix models presented in section 3.3.1. If an individual gamedesign feature obtains an F-Score of 1.0, it has the potential to drive the outcome of a game to success. However, the complexity ofmodern games means that more than one feature is typically found in a game. Therefore, it is important to explore featurecombinations, in order to quantify their impact on increasing or decreasing the F-Score metric. This combination could be the

12simultaneous implementation of two or more game design features in order to increase the probability of a game being successful.For example, to analyze the impact of a combination of features on the F-Score, the confusion matrix in Table 5 is presented.Table 5 Confusion Matrix for two game design features togetherClassVariablePositive OutcomeNegative Outcome(Game Design Feature 1) U (Game Design Feature 2)PresentAbsentTrue Positive(TP)False Positive(FP)False Negative(FN)True Negative(TN)Table 5 shows the confusion matrix for two game design features being analyzed together. This confusion matrix could be constructedfor more than two game design features. The confusion matrix will result in an F-Score for the selected game design features, basedon the precision and recall measures. However, selection of these game design features itself poses a computational problem becauseof the number of permutations generated based on the number of features. From an optimization perspective, the aim is to maximizethe F-Score by selecting an optimal combination of two or more game design features. For a set of n game design features, the totalnumber of permutations will depend of the number of features r considered at a time (equation 4).𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑟𝑚𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑠 𝑃 (𝑛, 𝑟) 𝑛!(𝑛 𝑟)!(4)It is essential to explore the impact of each feature on every other feature, as there may exist interaction effects between featuresshared by games (Filho et al., 2014). This process will terminate after obtaining the maximum F-Score possible, based on a givencombination of game design features. Therefore, as the number of game design features increases, it becomes difficult to sequentiallyanalyze the interactions between various feature combinations. Furthermore, it is difficult to determine the order in which thesefeature combinations should be evaluated, towards maximizing the F-Score. Hence, computationally efficient methods are necessaryto generate a model for analysis of these interactions. Machine learning algorithms have the ability to build relations between featureswhile computing the importance of each feature in the resulting predictive model. There are a number of machine learningclassification algorithms that can be employed to determine optimal feature combinations, relative to an output variable (i.e., in thiscase, successful or unsuccessful games). Table 6 presents a comparison of various classification algorithms, based on certain metrics.These algorithms have been shown to perform exceptionally well across a wide variety of classification tasks (Behoora and Tucker,2015). In Table 6, four stars represent the best performance attainable, while a one star represents the worst performance attainable.The SVM, Naïve Bayes, IBK, Decision Trees and Random Forest are presented in Table 6, based on a comparison presented by(Kotsiantis, 2007).

13Table 6 Comparison of Classifiers used (Kotsiantis, 2007), where **** represents the best and * represents the worst performance.AlgorithmsDecision Tree (C4.5, Random Forest)Naïve BayesSVMIBKModel Accuracy*********Explanation Ability of Classification***********Dealing with danger of overfittingTolerance to Independent Attributes*****************Classification algorithms are evaluated based on a variety of metrics such as accuracy, F-Score and relative error between the actualand predicted variables. Each of the above classification algorithms will be employed to efficiently discover relevant game designfeature combinations that predict the success or failure of games.4. Case StudyCasual gaming is considered to be a popular activity during leisure time and has gained in popularity, primarily due to thewidespread use

Some applications of gamification include enhancing employee engagement, creating healthy competition among teams, . gamification strategies helped patients recover from physical and mental wellness by enabling them to better adhere to their treatment regime (McCallum, 2012). . Marketing is another field where gamification concepts have .